Uncertainty propagation in a flood model cascade under different rainfall generation processes

Juan Rodriguez-Rincon, Jose Agustin Brena-aranjo, Adrian Pedrozo-Acuña

Tuesday 30 june 2015

15:20 - 15:35h at North America (level 0)

Themes: (T) Extreme events, natural variability and climate change, (ST) Hydrological extremes: floods and droughts

Parallel session: 6I. Extreme events - Flood Drought

Measured and numerical weather prediction (NWP) simulated rainfall products typically show differences in their spatial and temporal distribution. These differences can considerably influence the ability to predict hydrological responses. For flood inundation studies, it is desirable to implement a meteorological-hydrologic-hydraulic model chain to quantify the extent of flood risk. This requires a combination of modelling capabilities, the non-linear transformation of rainfall to river flow using rainfall-runoff models, and finally the hydraulic flood wave propagation based on the runoff predictions. The combination of these numerical tools involves the interaction of several sources of error that may affect an adequate characterisation of flood risk. However, both the propagation of uncertainties through the model chain, and how these errors affect the result under different meteorological conditions (e.g. tropical cyclone, cold front) are rarely examined. Therefore, the purpose of this investigation is to explore the effects of errors in rainfall prediction from an NWP, on inundation predictions for three different meteorological events that produced several damages in Mexico. The methodology is comprised of a Numerical Weather Prediction Model (NWP), a distributed rainfall-runoff model and a standard 2D hydrodynamic model. The cascade of models is implemented for two recent extreme flood events that took place in Mexico (Tonala, 2009 and Acapulco, 2013). In both cases, high quality field data (e.g. LiDAR; rain gauges) and satellite imagery are available. Uncertainty in the meteorological model (Weather Research and Forecasting model) is evaluated through the use of a multi-physics ensemble technique, which considers 25 parameterisation schemes to determine a given precipitation. The resulting precipitation fields are used as input in a distributed hydrological model, enabling the determination of different hydrographs associated to this event. Lastly, by means of a standard 2D hydrodynamic model, resulting hydrographs are used as forcing conditions to study the propagation of the meteorological uncertainty to an estimated flooded area. Differences in area and water level at benchmark stations are compared and the uncertainties at each modelling stage are analysed. Results show the utility of the selected modelling approach to investigate error propagation within a cascade of models.